I’ve been troubleshooting an issue where a single node would have all of its workers grow in memory until they would be OOM killed. The troubleshooting process spanned multiple days and I only happened to come across this with some help from Justin combined with a thread on the issue tracker (https://bro-tracker.atlassian.net/browse/BIT-1482).
Keep in mind that when you are using the MHR script (enabled by default) or the notary script, your Bro workers are performing a LOT of DNS. In my case I was using both. Since lookup_host_txt and lookup_host never return if the worker node doesn’t reach a DNS server, this results in what would appear to be a new thread for each new DNS query when your DNS resolvers are misconfigured.